Appearance
11.6 — Us and Them, and What Reduces It
Henri Tajfel wanted to find the minimum conditions required to produce group bias. He expected to have to build up gradually — shared history, competition, real stakes.
He never got that far.
Boys were sorted into two groups by a coin toss, or by which of two abstract paintings they said they preferred. No contact between groups. No history. No conflict. No shared interest. The group membership was explicitly arbitrary and everyone knew it.
Then they allocated money between an anonymous member of their own group and an anonymous member of the other.
They favoured their own group. And in a finding that is more disturbing than the favouritism, many chose options that gave their own group less in absolute terms, in order to increase the gap between the two.
This is the minimal group paradigm, and the result has been reproduced many times. The threshold for producing group bias is essentially zero.
Why it happens
Social identity theory, Tajfel's account: part of your self-concept comes from the groups you belong to, so raising your group's standing raises yours. The favouritism is self-esteem maintenance conducted at group level (5.6).
Two additional mechanisms, both well established.
Out-group homogeneity. You see enormous variety within your own group and much less within others. They are all the same; we are all individuals. This is partly a data problem — you know more members of your own group — and partly perceptual.
And the in-group is given the benefit of the doubt. The same act is interpreted charitably from a member of your group and uncharitably from a member of another. Ambiguity resolves in favour of your own.
What escalates it
Muzafer Sherif's Robbers Cave study, run at a summer camp, showed the sequence. Two groups of boys, initially unaware of each other, were introduced and put into competition for prizes.
Hostility appeared quickly — name-calling, raids, flag-burning, refusal to eat together.
Then Sherif tried to fix it. Simply bringing them together produced more conflict, not less. Contact alone made it worse.
What worked was a shared problem neither group could solve alone. The water supply broke and both groups had to work on it. A lorry needed to be pulled. After several such tasks, the hostility largely dissolved.
Caveats worth stating: it was a small study, run with a specific and unrepresentative group of boys, and Sherif's earlier attempt at the same design failed when the boys refused to become hostile. The findings are suggestive rather than definitive — and the shared-goal result is supported by a great deal of later work.
What actually reduces it
The best evidence here is on the contact hypothesis, originally set out by Gordon Allport: contact between groups reduces prejudice under specified conditions.
A large meta-analysis by Thomas Pettigrew and Linda Tropp, covering hundreds of studies and a quarter of a million people, found that intergroup contact reduces prejudice reliably, and that the effect appears even when Allport's conditions are not fully met — though it is stronger when they are.
Allport's four conditions:
Equal status within the situation. Contact where one group serves the other reinforces rather than reduces.
A shared goal. The Robbers Cave finding.
Cooperation rather than competition.
And institutional support — the arrangement being sanctioned by whoever is in authority.
What mediates it: the analysis found that reduced anxiety about the interaction and increased empathy account for much of the effect. The mechanism is largely that unfamiliarity produces anxiety, and contact removes the unfamiliarity.
And two further findings that make it practical.
Friendship is the strongest form. One genuine friendship across a group boundary does more than a great deal of casual contact.
And it generalises indirectly. Merely knowing that someone in your group has a friend in the other group reduces prejudice — the extended contact effect. Which matters, because it means the effect spreads beyond the people who actually met.
What does not work
Contact without the conditions. Competitive or unequal contact reinforces.
Telling people they are biased. Produces defensiveness, and there is evidence that some diversity training formats produce backlash rather than change.
Implicit bias training as a standalone intervention. The measure most commonly used — the Implicit Association Test — has known problems with test-retest reliability, and its ability to predict individual behaviour is weak and disputed. Meta-analyses find that changes in implicit measures do not reliably produce changes in behaviour. This does not mean prejudice is not real; it means this particular measurement and this particular intervention have been oversold.
And arguing. Almost nothing in the literature suggests that presenting counter-evidence to someone's group prejudice changes it, because the belief is identity-attached (1.5).
What does work, structurally
Changing what people do rather than what they think. Institutional rules — how hiring is done, who is in the room, who is on the shortlist — change behaviour immediately, and attitudes follow behaviour (9.8) rather than the other way round.
Superordinate identity. Getting people to see themselves as members of a single larger group. This works and it has a limit: it tends to work best when it does not require anyone to give up an identity they value, and it can fail when the larger identity is perceived as belonging to one of the groups already.
Individuating information. The bias operates on categories. Anything that makes a specific person specific — their name, their story, their situation — degrades the category. This is why 10.9's method is an anti-prejudice technique as well as a reading technique.
And exposure to counter-examples over time, particularly ones that cannot be dismissed as exceptional.
The version that operates in your ordinary life
This material is usually discussed in terms of ethnicity, religion and nationality, and the same mechanism runs at every scale.
Departments in an organisation. Sales and engineering. Clinical and administrative. Each believes it does the real work and the other creates the obstacles, and each has out-group homogeneity about the other.
Families in a marriage. His side and hers.
Teams, factions, schools of thought, the people who joined before and after.
And online, where group boundaries are explicit, contact is hostile, and the conditions Allport specified are absent by design.
The practical value of this page is largely at that scale, because it is the one where you can actually change something: get a shared problem, work on it cooperatively, and get one real friendship across the line.
What to do with this
Notice when you are using "they" about a group you belong to a rival of (10.4). It is a reliable marker that the category has taken over from the individuals.
Find the one person on the other side you can have a real conversation with. Not to debate. One genuine relationship across a boundary does more than every argument you will ever have, and the effect spreads to people who never met them.
Create a shared problem where you can. A joint task with a real deadline and a real outcome does more for two hostile departments than any amount of team-building.
Individuate deliberately. When you catch yourself thinking in a category, ask for one specific person's name and situation.
And check your own out-group homogeneity, which is the easiest of these to test. Can you name three ways in which members of the group you are annoyed with differ from each other? If not, you are not thinking about people.
Next: 11.7 covers the dimension underneath most group behaviour and most workplace conflict — status, and the difference between the two ways of getting it.